Circular Dependency.

When two Spring beans keep asking for each other first, the container gets stuck — interviewers love this because it tests both diagnosis and design judgment.

Question: What is a circular dependency in Spring Boot?

Answer: A circular dependency happens when bean A needs bean B, and bean B also needs bean A, so Spring cannot finish building either one cleanly. In Spring Boot 2.6 and later, this usually fails by default during startup with a BeanCurrentlyInCreationException. The safest fix is to break the loop by redesigning the beans; quick workarounds like @Lazy or ObjectProvider only defer the problem.

Interview-Ready Answer: In Spring Boot, a circular dependency means two or more beans depend on each other in a loop, like A → B → A. If the cycle is through constructors, Spring cannot create either bean, so startup fails; since Boot 2.6, that failure is the default behavior unless you explicitly change it. My first choice is to remove the cycle by refactoring, and if I need a temporary bridge, I use @Lazy or ObjectProvider to defer one side.

What it really means

A circular dependency is not just “two classes reference each other in code.” It becomes a Spring problem when both beans are needed during container startup. Spring must fully create bean A before bean B, but bean B also needs bean A, so the graph has no clean starting point.

How Spring handles it under the hood

1. Spring reads the bean graph and sees that one bean needs another, usually through constructor, setter, or field injection.
2. If bean A asks for bean B, Spring starts creating bean B.
3. If bean B then asks for bean A, Spring checks whether A can be exposed early. This only works for singleton beans and only in some injection styles.
4. For old-style singleton cycles with setter or field injection, Spring can publish an early reference from its singleton cache so the second bean can hold a temporary reference.
5. After that, Spring finishes property injection and initialization, then moves the bean into the final singleton cache.
6. If the cycle is through constructors, there is no half-built object to expose, so Spring stops with a startup failure. Boot 2.6+ makes this stricter by default with spring.main.allow-circular-references=false.

Memory detail: the “early reference” trick is a Spring-only safety net, not a real solution. It means one bean may see another bean before it is fully initialized, which is why circular references are considered a design smell.

Best ways to deal with it

Important note: @DependsOn only changes startup order; it does not remove the cycle. It can make the failure appear in a different place, but it does not solve the root problem.

When and why to use a workaround

• Use @Lazy when one side is rarely touched and you need a quick bridge during migration.
• Use ObjectProvider when the dependency is truly optional or only needed in one method.
• Use refactoring when both beans are core parts of the same business rule; that is usually the real fix.
• Avoid relying on the global Boot flag in new code; it can hide a design problem until the next upgrade.

Edge cases interviewers like

• Constructor injection: always the hardest to cycle-break because Spring must resolve everything immediately.
• Prototype beans: circular resolution is not the same as singleton resolution; early singleton exposure does not rescue you here.
• Lifecycle methods: a bean may inject fine but still create a runtime loop if @PostConstruct or a business method keeps calling back and forth.
• Proxy overhead: @Lazy adds one proxy hop, which is tiny in practice; the real cost is architectural complexity, not CPU time.

Version clue: Boot 2.5 and earlier generally allowed circular references by default; Boot 2.6 changed the default to false, so older apps suddenly started failing on upgrade. That is a classic interview trap.

Real-World Example: Imagine a checkout service in an e-commerce app. OrderService creates orders and calls PaymentService to charge the card. Later someone adds a “check order status” call inside PaymentService that reaches back into OrderService. In a local test, it looks harmless. In production after a Spring Boot 2.6 upgrade, the whole app fails to start, Kubernetes keeps restarting the pod, and the logs show an UnsatisfiedDependencyException ending in BeanCurrentlyInCreationException.

The user impact is brutal: checkout pages return 500s, no order can be placed, and every deploy is blocked until the cycle is removed. The fix is usually to extract a third service, such as OrderQueryService or PaymentPolicyService, so neither bean needs to know too much about the other. If the team needs a fast temporary release, one side can be marked @Lazy, but the long-term repair is to untangle the responsibility split.

What goes wrong when misunderstood: teams often think “just turn the flag on.” That may let the app boot, but it leaves a fragile design in place. The next refactor, proxy change, or scope change can break it again, and the outage shows up as startup failure, not a neat unit-test error.

Follow-up & Tricky Questions:

• How do you fix a circular dependency without changing the global Boot property? Break the design loop with a new service, or defer one side with @Lazy or ObjectProvider. That keeps the fix local instead of changing how the whole app behaves.
• Why does constructor injection fail more often than setter injection? Constructor injection needs every dependency up front, so there is no half-built object to expose. Setter or field injection can sometimes be satisfied later, but only in singleton scenarios and only when Spring is allowed to expose an early reference.
• What exception do you usually see? The top-level exception is often UnsatisfiedDependencyException, and the root cause is commonly BeanCurrentlyInCreationException. The root cause matters because it points to the bean loop, not just a missing bean.
• Is @Lazy a real fix? It is a useful workaround, not the best long-term architecture. It breaks the creation loop by inserting a proxy, but if the two classes truly depend on each other’s business logic, refactoring is cleaner.
• What should you do in a legacy app that already has many cycles? Turn the app back on with the least risky temporary workaround, then remove the cycles one by one. A slow cleanup is better than leaving the global flag on forever.
• Does @DependsOn solve it? No. It only changes order; it does not remove the dependency loop, so the container still has to resolve the same circular graph.
• Can circular dependencies happen at runtime after startup? Yes. A bean can be created successfully and still call back into the other bean in a loop during a method call or lifecycle callback, which causes recursion or repeated side effects.
• Is a prototype-scoped bean cycle treated the same way? No. The common singleton early-reference trick does not apply the same way, so prototype cycles are usually much harder or impossible to rescue with the usual Spring shortcuts.

• Tricky: If the app starts, does that mean there is no circular dependency? Not necessarily. Older Spring Boot apps, setter injection, or @Lazy can hide the cycle at startup while the design problem still exists.
• Tricky: Can I fix every cycle by setting allow-circular-references=true? No. That only relaxes startup behavior for some singleton cases; it does not make constructor cycles magically safe, and it should not be your long-term plan.
• Tricky: Is a bidirectional JPA relationship the same thing as a Spring bean cycle? No. That is a data model relationship, not a bean creation loop. The names sound similar, but the failure mode is different.

Common Mistakes:

• Mistake: Saying circular dependency is always fine if Spring can "figure it out." Correction: It may have worked in older versions for some singleton setter/field cases, but Boot 2.6+ fails by default, and constructor cycles still fail.
• Mistake: Recommending @DependsOn as the fix. Correction: It only changes initialization order; it does not break the loop.
• Mistake: Using allow-circular-references=true as the answer for new code. Correction: Treat it as a temporary migration switch, not a design strategy.
• Mistake: Forgetting that a successful startup can still hide a runtime loop. Correction: Watch for recursive business calls and @PostConstruct interactions too.

Memory Hook: Think of two people at a narrow door, each saying “you go first.” Nobody moves. Spring’s fix is either to open a side door (@Lazy) or redesign the hallway so they are not blocking each other.

Cheat Sheet:

• Circular dependency = bean A needs bean B and bean B needs bean A.
• Constructor cycles fail because there is no half-built object to hand out.
• Boot 2.6+ defaults to allow-circular-references=false.
• @Lazy breaks the loop with a proxy; ObjectProvider breaks it with late lookup.
• Refactoring is the best long-term fix; the flag is only a temporary escape hatch.
• Watch for startup exceptions like BeanCurrentlyInCreationException.

Practice Tasks:

• Remove @Lazy from the code example and observe the startup failure.
• Replace the direct call from PaymentService back to OrderService with a third service that owns the shared logic.
• Rewrite one dependency using ObjectProvider<PaymentService> instead of constructor injection.